User Guide¶

Comprehensive guides for using LangCalc in practice.

Contents¶

• Language Models - Working with different model types
• Algebraic Operations - Composing models mathematically
• Context Transformations - Transform context before prediction
• Examples & Patterns - Real-world usage patterns
• Best Practices - Tips and recommendations

Quick Navigation¶

By Use Case¶

• Text generation → Models + Algebra
• Pattern matching → Transformations
• Model mixing → Algebra + Examples
• Performance optimization → Best Practices

By Experience Level¶

• Beginners → Start with Models, then Examples
• Intermediate → Focus on Algebra and Transformations
• Advanced → See Best Practices and Advanced Topics

Common Workflows¶

Workflow 1: Building a Text Predictor¶

1. Load corpus and create infinigram model
2. Optionally compose with LLM
3. Add projections for robustness
4. Optimize based on use case

See Examples for details.

Workflow 2: Grounding an LLM¶

1. Create infinigram from knowledge base
2. Create or load LLM
3. Mix with optimal weights (95% LLM + 5% infinigram)
4. Evaluate perplexity improvement

See Examples for details.

Workflow 3: Custom Model Composition¶

1. Define component models
2. Choose algebraic operations
3. Tune weights experimentally
4. Add temperature scaling if needed

See Algebra for complete operator reference.